Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261005T121346Z-scsemi-cv-budget-interim/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/immicro/scsemi_cv_20261002/fold0/README.md

@@ -0,0 +1,60 @@
+# score_ft.py output: immicro, 22 scored donors
+
+Arms: scsemi__raw.  Every arm is scored on the same donors against the same observed reference.
+
+## Inputs and fixed contracts
+- Observed reference: `/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/immicro/reference.npz` (sha c5e2547be3ec), built from datadir:/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/immicro;
+  observed cells sum exactly to the features pseudobulk of every donor. Observed labels: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/immicro/labels/observed_immicro/labels.parquet.
+- Generated cells: `generated_counts` read out once with Poisson(seed 20260916) over the **whole NPZ in row order**
+  (`poisson_readout.poisson_integerise`), then the scored donors' rows. These integer counts are used for labels, embedding,
+  depth and the contrast. Pseudobulk uses the summed `generated_counts` (rates), as `score_generated.py`.
+- Labels: frozen f01common scANVI (`predict_scanvi.py`, CPU, zero query epochs), six-type gate ['T cell', 'B cell', 'Monocyte', 'NK cell', 'DC', 'Other']; query genes =
+  the 1,006-gene annotation list minus genes the dataset's sc assay did not measure in every sample (1004 genes).
+- Scoring genes: 6k panel genes measured by both assays in every sample of the dataset (5508 genes).
+
+## Definitions
+1. **Composition**: fraction of each donor's cells per six-type label (all generated cells; all observed QC cells).
+   `comp_l1` = sum over the six labels of |generated - observed| (immicro_common.l1). `comp_l1_obs_loo_null` = L1 of the
+   donor's observed composition to the mean of the *other scored donors'* observed compositions. `composition_top1` =
+   fraction of donors whose generated composition is closest (L1) to their own observed composition.
+2. **Pseudobulk**: `pb_r` = Pearson r of CPM-log1p(summed generated_counts) vs CPM-log1p(observed pseudobulk) on the scoring
+   genes (evaluate_realbulk.cpm_log1p / paired_r). `pb_top1` = the donor's generated pseudobulk is closest (Pearson) to its own
+   observed pseudobulk among the 22 scored donors (immicro_common.retrieval = evaluate_realbulk.top1 arithmetic).
+3. **Precision / coverage / density** (stage_b_embed.py functions): per-cell CP10K over the scoring genes + log1p; PCA-50
+   (randomized, random_state 20260925) and k = 10 real-real k-NN balls fitted on **R1 of the scored donors only**.
+   Reference rule (independent of fold and arm): per donor, cells in canonical order (library_id+barcode), one
+   permutation from `default_rng([20260925, uint64(sha256(donor)[:8])])`; R1 = first n, R2 = next n,
+   n = min(150, n_cells // 2) (3300 R1 cells here). Generated query: per donor n cells (the donor's R1 size),
+   without replacement, `default_rng([20260926, same key])`. Precision = fraction of generated cells inside any R1 ball
+   (Kynkaanniemi 2019); coverage = fraction of R1 cells whose ball holds a generated cell; density = Naeem 2020.
+   Per-donor precision = that donor's generated cells; per-donor coverage = that donor's R1 cells (the pooled values are
+   their n-weighted means; plain means when every donor has 150 reference cells). R2 (held-out real cells of the same donors) is the
+   perfect-generator control: precision 0.9958, coverage 1.0000.
+   **Depth matched**: stage_b_embed depth matching run pairwise (R1 vs this arm): per donor and within-donor rank of detected
+   genes, both cells are thinned by UMI sampling without replacement to the smaller detected-gene count; PCA kept from
+   un-thinned R1, radii recomputed on thinned R1; fresh rng(20260929) per arm. R2 control: precision
+   0.9936, coverage 1.0000.
+4. **Monocyte %** = 100 x Monocyte fraction (all cells) per donor; **median UMI per cell** = median over cells of the
+   Poisson-read (generated) / observed counts summed over the scoring genes (per donor, and pooled).
+5. **Group contrast** (T, B, Monocyte, NK): not computed (dataset immicro has no group contrast (use --groups-csv / --case-groups)).
+   Genes per type = scoring genes detected (count > 0) in >= 1 % of the scored donors' observed
+   cells of that type. Primary statistic `mean_cp10k`: per-cell CP10K over the scoring genes; per donor the mean over its
+   cells of the type (donors with >= 1 such cell); group mean = mean over the group's donors (donor-balanced);
+   logFC = log2(case mean + 1) - log2(control mean + 1). Secondary `pseudobulk_log1p_cpm`: the SLE benchmark evaluator's
+   statistic (donor x type summed counts, log1p CPM over the scoring genes, mean case - mean control;
+   evaluate_generated_vs_truth.group_log_fold_change). Both are computed on the generated cells (their frozen labels) and on
+   the observed cells; `r_generated` = Pearson r of the two logFC vectors. Broadcast baseline `r_broadcast`: the same
+   observed vector against the logFC of the input bulk (condition NPZ `bulk_counts` of the arm, scoring genes, per-donor
+   CP10K resp. log1p CPM, group means). `partial_r_generated_given_broadcast`: r of the OLS residuals of observed and
+   generated logFC on the bulk logFC (residual_specificity.py).
+6. **Top-50 DEG overlap** (contrast.csv rows `statistic = welch_top50_deg`; summary `deg_top50`): not computed (contrast off).
+
+## Files
+- `per_donor.csv`: one row per arm x donor (all per-donor quantities above).
+- `summary.json`: per arm the donor means / pooled values, provenance (NPZ sha, receipt check, generation metadata,
+  Poisson nnz of the whole NPZ, annotation contract), and the reference block (PCA, R2 control).
+- `contrast.csv`: per arm x cell type x statistic (`mean_cp10k`, `pseudobulk_log1p_cpm`: logFC r; `welch_top50_deg`:
+  top-50 DEG overlap, T / B / Monocyte / NK / DC; columns of the other kind are empty).
+- `arms/<arm>/labels/` (predict_scanvi.py labels + manifest), `arms/<arm>/generated_cells.parquet` (NPZ row, donor, label,
+  UMI on the scoring genes, P/C subsample flag), `arms/<arm>/deg_top50_lists.csv` (observed, generated and broadcast
+  top-50 gene lists, Ensembl ids, rank order).

paper-releases/pbisc-diffusion/updates/20261005T121346Z-scsemi-cv-budget-interim/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/immicro/scsemi_cv_20261002/fold0/per_donor.csv

@@ -0,0 +1,23 @@
+arm,donor,group,n_generated_cells,n_observed_cells,n_reference_cells,gen_T cell,obs_T cell,gen_B cell,obs_B cell,gen_Monocyte,obs_Monocyte,gen_NK cell,obs_NK cell,gen_DC,obs_DC,gen_Other,obs_Other,comp_l1,comp_l1_obs_loo_null,pb_r,pb_own_rank,pb_top1,precision,coverage,precision_depth_matched,coverage_depth_matched,mono_pct_gen,mono_pct_obs,median_umi_gen,median_umi_obs
+scsemi__raw,HS3,DB3,1500,4617,150,0.712,0.6484730344379467,0.030666666666666665,0.06952566601689408,0.13733333333333334,0.17782109594975093,0.03933333333333333,0.07060862031622266,0.076,0.028589993502274202,0.004666666666666667,0.004981589776911414,0.22187394411955821,0.04060118700472884,0.707553614163355,1,True,0.68,0.08666666666666667,0.88,0.4,13.733333333333334,17.782109594975093,1957.0,2186.0
+scsemi__raw,HS6,DB3,1500,4222,150,0.5666666666666667,0.4836570345807674,0.08733333333333333,0.037423022264329704,0.25866666666666666,0.3147797252486973,0.036,0.13003315963998105,0.05,0.024869729985788725,0.0013333333333333333,0.009237328280435813,0.31610042633822827,0.3824588765473767,0.664976909894266,4,False,0.7133333333333334,0.06,0.84,0.44,25.866666666666667,31.47797252486973,1955.5,2457.5
+scsemi__raw,HS7,DB1,1500,4271,150,0.702,0.7035822992273473,0.056666666666666664,0.03839850152189183,0.11533333333333333,0.13135097166939827,0.07933333333333334,0.09740107703114025,0.03933333333333333,0.02130648560056193,0.007333333333333333,0.0079606649496605,0.07259002575509256,0.17462025766050882,0.6510496747817035,2,False,0.7466666666666667,0.12666666666666668,0.8466666666666667,0.4066666666666667,11.533333333333333,13.135097166939827,1957.0,1748.0
+scsemi__raw,HS13,DB5,1500,4792,150,0.6953333333333334,0.6156093489148581,0.042666666666666665,0.056343906510851416,0.13466666666666666,0.2166110183639399,0.04733333333333333,0.08764607679465776,0.07066666666666667,0.02003338898163606,0.009333333333333334,0.0037562604340567614,0.27186867000556486,0.07950263318013584,0.6359461936637109,18,False,0.7266666666666667,0.05333333333333334,0.8266666666666667,0.26666666666666666,13.466666666666665,21.66110183639399,1958.0,2176.0
+scsemi__raw,HS18,DB4,1500,3348,150,0.7026666666666667,0.6633811230585425,0.04066666666666666,0.05734767025089606,0.13933333333333334,0.18309438470728792,0.04,0.07198327359617682,0.062,0.016129032258064516,0.015333333333333332,0.008064516129032258,0.18485065710872153,0.056830454472214534,0.6641027584341265,14,False,0.7133333333333334,0.05333333333333334,0.7266666666666667,0.38,13.933333333333334,18.309438470728793,1958.0,2108.0
+scsemi__raw,HS31,DB3,1500,4830,150,0.6406666666666667,0.5956521739130435,0.08533333333333333,0.07039337474120083,0.138,0.21884057971014492,0.09333333333333334,0.08385093167701864,0.038,0.02546583850931677,0.004666666666666667,0.005797101449275362,0.16394202898550725,0.08982859354260735,0.7590017764119595,1,True,0.8066666666666666,0.04666666666666667,0.86,0.38,13.8,21.884057971014492,1955.0,2217.5
+scsemi__raw,HS34,DB5,1500,4363,150,0.728,0.7073114829245931,0.08133333333333333,0.04859041943616777,0.098,0.16983726793490717,0.05466666666666667,0.0550080220032088,0.03,0.013981205592482237,0.008,0.005271602108640843,0.14435724654289864,0.14554772243505307,0.6513960064049729,19,False,0.7466666666666667,0.03333333333333333,0.76,0.24,9.8,16.983726793490717,1955.5,2176.0
+scsemi__raw,HS37,DB6,1500,2127,150,0.5953333333333334,0.5763986835919135,0.058666666666666666,0.14997649271274094,0.202,0.18288669487541137,0.10333333333333333,0.06535025858015985,0.03133333333333333,0.021626704278326282,0.009333333333333334,0.0037611659614480487,0.18261965209214864,0.17642904039178112,0.6860776798535069,13,False,0.88,0.06,0.78,0.38,20.200000000000003,18.28866948754114,1955.0,2099.0
+scsemi__raw,HS38,DB2,1500,3952,150,0.556,0.4534412955465587,0.07066666666666667,0.06022267206477733,0.154,0.27631578947368424,0.18,0.16776315789473684,0.03666666666666667,0.03365384615384615,0.0026666666666666666,0.00860323886639676,0.2565047233468287,0.3979973492276381,0.7319161670129157,2,False,0.8266666666666667,0.25333333333333335,0.8,0.32666666666666666,15.4,27.631578947368425,1955.0,1448.0
+scsemi__raw,HS52,DB1,1500,4286,150,0.65,0.7473168455436304,0.062,0.04316378908072795,0.16266666666666665,0.12085860942603827,0.06466666666666666,0.06276248250116659,0.056666666666666664,0.017498833411105925,0.004,0.008399440037330844,0.20343257116192248,0.23339752310597672,0.650706418783034,8,False,0.6866666666666666,0.14,0.84,0.4533333333333333,16.266666666666666,12.085860942603826,1957.0,1848.5
+scsemi__raw,HS58,DB7,1500,1299,150,0.748,0.6605080831408776,0.044,0.029253271747498075,0.148,0.15550423402617397,0.012,0.1170130869899923,0.04,0.03156274056966898,0.008,0.006158583525789068,0.22503464203233253,0.13904887966629118,0.6291653602712789,18,False,0.7866666666666666,0.04666666666666667,0.8,0.28,14.799999999999999,15.550423402617398,1957.5,1944.0
+scsemi__raw,HS60,DB6,1500,2659,150,0.708,0.6288078224896577,0.07266666666666667,0.04061677322301617,0.12866666666666668,0.16058668672433246,0.042,0.14328694998119593,0.04733333333333333,0.020308386611508085,0.0013333333333333333,0.006393380970289583,0.2765340353516359,0.12991818939150382,0.6899068467343682,7,False,0.7733333333333333,0.08,0.8266666666666667,0.31333333333333335,12.866666666666667,16.058668672433246,1955.0,2014.0
+scsemi__raw,HS66,DB4,1500,4857,150,0.7286666666666667,0.6448424953675108,0.082,0.10376775787523163,0.092,0.16512250360304714,0.026,0.07020794729256744,0.059333333333333335,0.012559192917438749,0.012,0.0035001029442042414,0.2781964175416924,0.09427105876197542,0.5868825447200179,18,False,0.7466666666666667,0.04,0.9666666666666667,0.17333333333333334,9.2,16.512250360304716,1955.0,2041.0
+scsemi__raw,HS81,DB7,1500,1638,150,0.7233333333333334,0.721001221001221,0.044,0.04212454212454213,0.12866666666666668,0.13553113553113552,0.04,0.06288156288156288,0.05733333333333333,0.03296703296703297,0.006666666666666667,0.005494505494505495,0.05949206349206352,0.19387054279631524,0.6453658981821437,14,False,0.6,0.04666666666666667,0.7666666666666667,0.3,12.866666666666667,13.553113553113553,1959.0,2240.5
+scsemi__raw,HS82,DB2,1500,4201,150,0.7006666666666667,0.6553201618662223,0.078,0.1309212092358962,0.12066666666666667,0.14782194715543917,0.05333333333333334,0.031183051654368008,0.04066666666666666,0.02499404903594382,0.006666666666666667,0.009759581052130446,0.16633880822026495,0.18293033059828231,0.6604352106943668,1,True,0.7733333333333333,0.23333333333333334,0.8666666666666667,0.4266666666666667,12.066666666666666,14.782194715543916,1957.0,1587.0
+scsemi__raw,HS83,DB6,1500,2410,150,0.6666666666666666,0.7128630705394191,0.088,0.08049792531120332,0.12,0.14481327800829877,0.066,0.045228215767634854,0.042,0.012448132780082987,0.017333333333333333,0.004149377593360996,0.14201936376210242,0.1880345194642001,0.6590244507904656,4,False,0.8066666666666666,0.03333333333333333,0.84,0.4533333333333333,12.0,14.481327800829877,1955.0,2373.0
+scsemi__raw,HS85,DB5,1500,4216,150,0.7073333333333334,0.5785104364326376,0.06266666666666666,0.07756166982922201,0.13866666666666666,0.22936432637571158,0.024,0.0896584440227704,0.06266666666666666,0.021347248576850095,0.004666666666666667,0.003557874762808349,0.3425022137887414,0.13514298327096805,0.5822616005591681,22,False,0.72,0.03333333333333333,0.98,0.16666666666666666,13.866666666666665,22.93643263757116,1958.0,2171.5
+scsemi__raw,HS86,DB7,1500,1111,150,0.7413333333333333,0.6561656165616562,0.04,0.0711071107110711,0.12333333333333334,0.162016201620162,0.012,0.0711071107110711,0.082,0.034203420342034205,0.0013333333333333333,0.0054005400540054005,0.2659285928592859,0.07179394704932163,0.6324078667329198,14,False,0.9,0.06,0.6733333333333333,0.48,12.333333333333334,16.2016201620162,1958.0,1989.0
+scsemi__raw,HS95,DB1,1500,4264,150,0.644,0.6458724202626641,0.07866666666666666,0.08348968105065666,0.12266666666666666,0.13860225140712945,0.05733333333333333,0.09826454033771106,0.08866666666666667,0.025093808630393996,0.008666666666666666,0.008677298311444653,0.12714571607254527,0.09728161934908136,0.6205233048338242,7,False,0.7333333333333333,0.07333333333333333,0.8866666666666667,0.44666666666666666,12.266666666666666,13.860225140712945,1957.0,1829.5
+scsemi__raw,HS98,DB2,1500,4959,150,0.5713333333333334,0.5366001209921355,0.06466666666666666,0.07360354910264166,0.21333333333333335,0.30490018148820325,0.06466666666666666,0.03891913692276669,0.07933333333333334,0.03367614438394838,0.006666666666666667,0.012300867110304496,0.21227586206896554,0.3008904548210157,0.6804188111949028,1,True,0.82,0.20666666666666667,0.8533333333333334,0.44666666666666666,21.333333333333336,30.490018148820326,1959.0,1606.0
+scsemi__raw,HS101,DB4,1500,6549,150,0.648,0.6659031913269201,0.084,0.051152847763017255,0.12866666666666668,0.192090395480226,0.10133333333333333,0.05939838143227974,0.037333333333333336,0.026721636891128418,0.0006666666666666666,0.004733547106428463,0.17078760116048247,0.08013023802852969,0.6582863310025578,15,False,0.7866666666666666,0.06,0.8466666666666667,0.26,12.866666666666667,19.2090395480226,1955.0,1960.0
+scsemi__raw,HS107,DB1,1500,4445,150,0.6906666666666667,0.7313835770528684,0.06933333333333333,0.031496062992125984,0.12266666666666666,0.14195725534308212,0.07666666666666666,0.0686164229471316,0.037333333333333336,0.02002249718785152,0.0033333333333333335,0.006524184476940383,0.12639670041244846,0.19608442509403828,0.6768473198714373,2,False,0.76,0.13333333333333333,0.7133333333333334,0.6,12.266666666666666,14.195725534308213,1957.0,1849.0

paper-releases/pbisc-diffusion/updates/20261005T121346Z-scsemi-cv-budget-interim/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/immicro/scsemi_cv_20261002/fold0/summary.json

@@ -0,0 +1,208 @@
+{
+ "schema": "paired_ft_scores.v1",
+ "dataset": "immicro",
+ "written_utc": "2026-10-05T12:05:21.591937+00:00",
+ "command": [
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/score_ft.py",
+  "--dataset",
+  "immicro",
+  "--out",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scores/immicro/scsemi_cv_20261002/fold0",
+  "--threads",
+  "8",
+  "--arm",
+  "scsemi__raw=/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/generated/immicro/fold0/scsemi__raw.npz"
+ ],
+ "arms": [
+  "scsemi__raw"
+ ],
+ "donors": [
+  "HS3",
+  "HS6",
+  "HS7",
+  "HS13",
+  "HS18",
+  "HS31",
+  "HS34",
+  "HS37",
+  "HS38",
+  "HS52",
+  "HS58",
+  "HS60",
+  "HS66",
+  "HS81",
+  "HS82",
+  "HS83",
+  "HS85",
+  "HS86",
+  "HS95",
+  "HS98",
+  "HS101",
+  "HS107"
+ ],
+ "n_donors": 22,
+ "reference": {
+  "cache": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/immicro",
+  "reference_sha256": "c5e2547be3ec3b12cb0c0b404a49e14a046ae59d40be70427f98f083a76bce17",
+  "features_sha256": "735255066ce7ebbac13a1fdfd058e26f31edd60314cc02fd589b0a292dbbc838",
+  "n_r1_cells": 3300,
+  "pca": {
+   "n_components": 50,
+   "explained_variance_ratio_sum": 0.3059765696525574,
+   "n_fit_cells": 3300
+  },
+  "r2_control": {
+   "precision": 0.9957575757575757,
+   "coverage": 1.0,
+   "density": 2.608393939393939,
+   "n_query": 3300,
+   "depth_matched": {
+    "precision": 0.9936363636363637,
+    "coverage": 1.0,
+    "density": 2.558878787878788
+   }
+  },
+  "observed_l1_loo_null_mean": 0.16302776481179745
+ },
+ "contrast_definition": {
+  "enabled": false,
+  "reason": "dataset immicro has no group contrast (use --groups-csv / --case-groups)"
+ },
+ "deg_definition": {
+  "enabled": false,
+  "reason": "contrast off"
+ },
+ "headline": {
+  "scsemi__raw": {
+   "comp_l1": 0.2004905437372287,
+   "comp_l1_obs_loo_null": 0.16302776481179745,
+   "precision": 0.7606060606060606,
+   "coverage": 0.0890909090909091,
+   "precision_dm": 0.8263636363636364,
+   "coverage_dm": 0.36454545454545456,
+   "pb_r_median": 0.6586553908965117,
+   "pb_top1": 0.18181818181818182,
+   "mono_pct_gen": 14.215151515151511,
+   "mono_pct_obs": 18.503211518737277,
+   "median_umi_gen": 1957.0,
+   "median_umi_obs": 1976.0,
+   "seconds": 7.3
+  }
+ },
+ "per_arm": {
+  "scsemi__raw": {
+   "arm": "scsemi__raw",
+   "npz": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/generated/immicro/fold0/scsemi__raw.npz",
+   "npz_sha256": "eeb6621c1cb8662742bc2cea3383f001dc3ae5df8697a7c10d03de29cdcf6cd7",
+   "receipt_verified": true,
+   "generation": {
+    "checkpoint": null,
+    "checkpoint_sha256": null,
+    "input_condition_path": "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/immicro/fold0/test_raw.npz",
+    "input_condition_sha256": "51ad5b8f668b55d80d2f8d275a3a2a9b56f7daa768d6b607dcea4b7379465f60",
+    "seed": 20260927,
+    "steps": null,
+    "support_readout": "upstream (values <= 10 normalized units zeroed by fast_semi)",
+    "weights": null
+   },
+   "condition_for_broadcast": null,
+   "poisson": {
+    "seed": 20260916,
+    "rule": "whole NPZ, rows in NPZ order",
+    "nnz_whole_npz": 4203237,
+    "median_library_whole_npz_panel": 1957.0,
+    "n_rows_whole_npz": 33000
+   },
+   "annotation": {
+    "model_pt_sha256": "bf3655899e5abee435340fc008656c4a17342da0f30b70cfd43de8e9223e1930",
+    "zero_padded_model_genes": 1,
+    "label_counts": {
+     "T cell": 22240,
+     "Monocyte": 4691,
+     "B cell": 2136,
+     "NK cell": 1970,
+     "DC": 1747,
+     "Other": 216
+    },
+    "mean_l2_max_prob": 0.917225239008665,
+    "predict_seconds": 3.7
+   },
+   "n_donors": 22,
+   "n_generated_cells": 33000,
+   "composition": {
+    "l1_mean": 0.2004905437372287,
+    "l1_median": 0.194141614135322,
+    "l1_obs_loo_null_mean": 0.16302776481179745,
+    "l1_generated_loo_mean_vs_observed_mean": 0.1911419578183098,
+    "composition_top1": 0.13636363636363635,
+    "mean_fraction_generated": {
+     "T cell": 0.673939,
+     "B cell": 0.064727,
+     "Monocyte": 0.142152,
+     "NK cell": 0.059697,
+     "DC": 0.052939,
+     "Other": 0.006545
+    },
+    "mean_fraction_observed": {
+     "T cell": 0.637846,
+     "B cell": 0.065772,
+     "Monocyte": 0.185032,
+     "NK cell": 0.081281,
+     "DC": 0.023594,
+     "Other": 0.006476
+    }
+   },
+   "pseudobulk": {
+    "n_genes": 5508,
+    "r_median": 0.6586553908965117,
+    "r_mean": 0.6601933065905,
+    "top1": 0.18181818181818182,
+    "n_top1": 4,
+    "own_rank_median": 7.5,
+    "chance_top1": 0.045454545454545456
+   },
+   "precision_coverage": {
+    "precision": 0.7606060606060606,
+    "coverage": 0.0890909090909091,
+    "density": 0.2569393939393939,
+    "n_query": 3300,
+    "n_reference": 3300
+   },
+   "precision_coverage_depth_matched": {
+    "precision": 0.8263636363636364,
+    "coverage": 0.36454545454545456,
+    "density": 0.5949393939393939,
+    "detected_median_before": {
+     "R1": 365.0,
+     "Q": 123.0
+    },
+    "detected_median_after": {
+     "R1": 123.0,
+     "Q": 123.0
+    }
+   },
+   "monocyte_pct": {
+    "generated_mean_over_donors": 14.215151515151511,
+    "observed_mean_over_donors": 18.503211518737277,
+    "generated_pooled_cells": 14.215151515151515
+   },
+   "median_umi_per_cell_scoring_genes": {
+    "generated_pooled": 1957.0,
+    "observed_pooled": 1976.0,
+    "generated_mean_of_donor_medians": 1956.6136474609375,
+    "observed_mean_of_donor_medians": 2003.1363636363637
+   },
+   "seconds": 7.3
+  }
+ },
+ "code_sha256": {
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/score_ft.py": "ec5f40b4cea6d5062d0138f3b934abb6465e34cbfeac869c8a0906924c033aec",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/tools/ft_common.py": "4a6366b0f22208cd58714fc0261b8840c643b9d4569704d560cfd93534cbc747",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/umap/cell_immicro_5arm/tools/stage_b_embed.py": "d056cfdf38fb0e5f478756f370174e993b75d0c973c48e02e21749175a4ca2da",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/bulk2sc-paper/scripts/evaluate_generated_vs_truth.py": "48b12d794782b840838853e93e0fc55fab00532c4b3698f4588e926b985d2881",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/immicro/tools/immicro_common.py": "c5c30e917148cc5e664e91ea30c11bbdff9b6d1f53a3e0a13e16b1917eff37e9",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/predict_scanvi.py": "482ee5dcf4d72b54bc36cbb6dcff62e5030bc7afc46cb5fb01eca73a82ee45a5",
+  "/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/aifi_scanvi_rebuild_20260916/poisson_readout.py": "8bfa080942fbcdc89ae23b276524a1a83266fb8cbc56b9af279e50f82e6cde38"
+ },
+ "seconds": 9.1
+}

paper-releases/pbisc-diffusion/updates/20261005T121346Z-scsemi-cv-budget-interim/diffusion/dataset/server_1/model8/plans/paired_finetune_intradataset_20260927/scores/immicro/scsemi_cv_20261002/fold1/README.md

@@ -0,0 +1,60 @@
+# score_ft.py output: immicro, 22 scored donors
+
+Arms: scsemi__raw.  Every arm is scored on the same donors against the same observed reference.
+
+## Inputs and fixed contracts
+- Observed reference: `/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/immicro/reference.npz` (sha c5e2547be3ec), built from datadir:/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/immicro;
+  observed cells sum exactly to the features pseudobulk of every donor. Observed labels: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/immicro/labels/observed_immicro/labels.parquet.
+- Generated cells: `generated_counts` read out once with Poisson(seed 20260916) over the **whole NPZ in row order**
+  (`poisson_readout.poisson_integerise`), then the scored donors' rows. These integer counts are used for labels, embedding,
+  depth and the contrast. Pseudobulk uses the summed `generated_counts` (rates), as `score_generated.py`.
+- Labels: frozen f01common scANVI (`predict_scanvi.py`, CPU, zero query epochs), six-type gate ['T cell', 'B cell', 'Monocyte', 'NK cell', 'DC', 'Other']; query genes =
+  the 1,006-gene annotation list minus genes the dataset's sc assay did not measure in every sample (1004 genes).
+- Scoring genes: 6k panel genes measured by both assays in every sample of the dataset (5508 genes).
+
+## Definitions
+1. **Composition**: fraction of each donor's cells per six-type label (all generated cells; all observed QC cells).
+   `comp_l1` = sum over the six labels of |generated - observed| (immicro_common.l1). `comp_l1_obs_loo_null` = L1 of the
+   donor's observed composition to the mean of the *other scored donors'* observed compositions. `composition_top1` =
+   fraction of donors whose generated composition is closest (L1) to their own observed composition.
+2. **Pseudobulk**: `pb_r` = Pearson r of CPM-log1p(summed generated_counts) vs CPM-log1p(observed pseudobulk) on the scoring
+   genes (evaluate_realbulk.cpm_log1p / paired_r). `pb_top1` = the donor's generated pseudobulk is closest (Pearson) to its own
+   observed pseudobulk among the 22 scored donors (immicro_common.retrieval = evaluate_realbulk.top1 arithmetic).
+3. **Precision / coverage / density** (stage_b_embed.py functions): per-cell CP10K over the scoring genes + log1p; PCA-50
+   (randomized, random_state 20260925) and k = 10 real-real k-NN balls fitted on **R1 of the scored donors only**.
+   Reference rule (independent of fold and arm): per donor, cells in canonical order (library_id+barcode), one
+   permutation from `default_rng([20260925, uint64(sha256(donor)[:8])])`; R1 = first n, R2 = next n,
+   n = min(150, n_cells // 2) (3300 R1 cells here). Generated query: per donor n cells (the donor's R1 size),
+   without replacement, `default_rng([20260926, same key])`. Precision = fraction of generated cells inside any R1 ball
+   (Kynkaanniemi 2019); coverage = fraction of R1 cells whose ball holds a generated cell; density = Naeem 2020.
+   Per-donor precision = that donor's generated cells; per-donor coverage = that donor's R1 cells (the pooled values are
+   their n-weighted means; plain means when every donor has 150 reference cells). R2 (held-out real cells of the same donors) is the
+   perfect-generator control: precision 0.9955, coverage 1.0000.
+   **Depth matched**: stage_b_embed depth matching run pairwise (R1 vs this arm): per donor and within-donor rank of detected
+   genes, both cells are thinned by UMI sampling without replacement to the smaller detected-gene count; PCA kept from
+   un-thinned R1, radii recomputed on thinned R1; fresh rng(20260929) per arm. R2 control: precision
+   0.9942, coverage 1.0000.
+4. **Monocyte %** = 100 x Monocyte fraction (all cells) per donor; **median UMI per cell** = median over cells of the
+   Poisson-read (generated) / observed counts summed over the scoring genes (per donor, and pooled).
+5. **Group contrast** (T, B, Monocyte, NK): not computed (dataset immicro has no group contrast (use --groups-csv / --case-groups)).
+   Genes per type = scoring genes detected (count > 0) in >= 1 % of the scored donors' observed
+   cells of that type. Primary statistic `mean_cp10k`: per-cell CP10K over the scoring genes; per donor the mean over its
+   cells of the type (donors with >= 1 such cell); group mean = mean over the group's donors (donor-balanced);
+   logFC = log2(case mean + 1) - log2(control mean + 1). Secondary `pseudobulk_log1p_cpm`: the SLE benchmark evaluator's
+   statistic (donor x type summed counts, log1p CPM over the scoring genes, mean case - mean control;
+   evaluate_generated_vs_truth.group_log_fold_change). Both are computed on the generated cells (their frozen labels) and on
+   the observed cells; `r_generated` = Pearson r of the two logFC vectors. Broadcast baseline `r_broadcast`: the same
+   observed vector against the logFC of the input bulk (condition NPZ `bulk_counts` of the arm, scoring genes, per-donor
+   CP10K resp. log1p CPM, group means). `partial_r_generated_given_broadcast`: r of the OLS residuals of observed and
+   generated logFC on the bulk logFC (residual_specificity.py).
+6. **Top-50 DEG overlap** (contrast.csv rows `statistic = welch_top50_deg`; summary `deg_top50`): not computed (contrast off).
+
+## Files
+- `per_donor.csv`: one row per arm x donor (all per-donor quantities above).
+- `summary.json`: per arm the donor means / pooled values, provenance (NPZ sha, receipt check, generation metadata,
+  Poisson nnz of the whole NPZ, annotation contract), and the reference block (PCA, R2 control).
+- `contrast.csv`: per arm x cell type x statistic (`mean_cp10k`, `pseudobulk_log1p_cpm`: logFC r; `welch_top50_deg`:
+  top-50 DEG overlap, T / B / Monocyte / NK / DC; columns of the other kind are empty).
+- `arms/<arm>/labels/` (predict_scanvi.py labels + manifest), `arms/<arm>/generated_cells.parquet` (NPZ row, donor, label,
+  UMI on the scoring genes, P/C subsample flag), `arms/<arm>/deg_top50_lists.csv` (observed, generated and broadcast
+  top-50 gene lists, Ensembl ids, rank order).